While I am at it, this patch uses const l-value references for
std::shared_ptr.  We don't need to increment the reference count by
passing std::shared_ptr by value.

Identified with llvm-use-ranges.

Identified with modernize-use-equals-default.

Identified with readability-redundant-typename.

The enum label e.g. `AK_arm64` is often the same as the architecture
name, but doesn't have to be. Seperate it out.

The SymbolSet class bump-ptr-allocates Symbol objects, but Symbol has a
non-trivial destructor (since Symbol's Targets member is a SmallVector):
we need to explicitly destroy the Symbol objects to ensure that no
memory is leaked.

rdar://154778728

resolves: rdar://151765704

- try_emplace(Key) is shorter than insert({Key, nullptr}).
- try_emplace performs value initialization without value parameters.
- We overwrite values on successful insertion anyway.

* Translate the following versions to 26.
  * watchOS 12 -> 26
  * visionOS 3 -> 26
  * macos 16 -> 26
  * iOS 19 -> 26
  * tvOS 19 -> 26

* Emit diagnostics, but allow conversion when clients attempt to use
invalid gaps in OS versioning in availability.

* For target-triples, only allow "valid" versions for implicit
conversions.

These are identified by misc-include-cleaner.  I've filtered out those
that break builds.  Also, I'm staying away from llvm-config.h,
config.h, and Compiler.h, which likely cause platform- or
compiler-specific build failures.

try_emplace value-initializes values, so we do not need to pass
nullptr to try_emplace when the value types are raw pointers or
std::unique_ptr<T>.